AI Agents on Joe Bollen Security

Improving a Coding Agent Harness: Part 1.5, Securely Reading Code

Tue, 07 Apr 2026 10:00:00 +0000

In Part 1, I added tree-sitter-based code understanding tools to mini-coding-agent, an open-source coding agent harness. The tools let the agent find definitions, trace references, and read specific functions by name instead of relying on grep and line-range reads. The code reading tools had no boundary enforcement, so the agent could read files outside the current directory and access sensitive files without restriction.

Improving a Coding Agent Harness: Part 1, Reading Code

Tue, 07 Apr 2026 09:00:00 +0000

Sebastian Raschka’s article on coding agent harnesses is a good reference for concepts that are easy to take for granted once you work with them daily. He breaks a coding agent into six components (live repo context, prompt shape and cache reuse, structured tools, context reduction, session memory, and delegation) and walks through each one with his mini-coding-agent, a clean, minimal implementation in about 1,000 lines of Python.